Test-Time Adaptation of Manipulation Policies Under Actuator Degradation
Test-Time Adaptation of Manipulation Policies Under Actuator Degradation
This research introduces TeAR, a policy-agnostic method that adapts manipulation policies in real-time using telemetry data to account for actuator degradation. Evaluated across 18 policy-task pairs, TeAR improves success rates by 10-15% under heating without requiring on-robot fine-tuning.
Evidence and limits
Published automatically after robotics and source-evidence checks; no manual editorial approval is recorded. Source assertions are not independently verified. Missing information remains not reported.
- Environment:
- Not reported
- Control:
- Not reported
- Data origin:
- Not reported
| Metric | Value / unit | Basis / context | Evidence |
|---|---|---|---|
| success | 31.8 percent | Reported trials Source wording: “31.8% success” | Source E1 |
| success | 25.6 percent | Reported trials Source wording: “25.6% for the base policy” | Source E1 |
| success | 30.6 percent | Reported trials Source wording: “30.6% for an assumed-model inverse” | Source E1 |
| improvement | 10 percent | Reported trials Source wording: “improves success under heating by 10-15%” | Source E1 |
| improvement | 15 percent | Reported trials Source wording: “improves success under heating by 10-15%” | Source E1 |
- dataset: Not reported
Source excerpts and review record
No manual editorial approval recorded.
Original source quotation: “TeAR achieves 31.8% success, compared with 25.6% for the base policy and 30.6% for an assumed-model inverse.”
Source E1
Source：arXiv Robotics — research abstracts · arxiv.org